Background understanding in video analytics applications refers to modeling the background of video image data, the static (non-moving) part of a static camera field-of-view that defines or otherwise belongs to the environmental settings of the image. For example, the road that moving objects travel upon, a building that moving pedestrians enter and exit, etc. By knowing the background and performing a difference operation between current video frame images with the background, moving foreground objects may be detected and identified, enabling other video analytics operations appropriate for the detected object, such as moving object tracking, classification, activity significance recognition and determination, etc.